Learning by (virtually) doing: Experimentation and belief updating in smallholder agriculture

Abstract

In much of sub-Saharan Africa, soil quality heterogeneity hampers farmer learning about the returns to different inputs. This may help explain why relatively few farmers in the region use improved inputs. We study how Kenyan farmers respond to an interactive app that enables them to discover the returns to different inputs on a virtual farm that is calibrated to resemble their own. Farmers update both their beliefs and their behaviors after engaging with the virtual learning app. We measure beliefs by eliciting probability distributions and use an incentive-compatible experiment to measure behavior change. The experiment gave participants an input budget that they could allocate across farm inputs. After playing several virtual seasons on the app, they could update these allocations. Farmers revise their input allocations along several dimensions after the virtual learning experience. To support our interpretation that these adjustments stem from real learning, we show that farmers with the highest predicted returns to lime—an unfamiliar input in this region—increase their lime orders more than others. Our results suggest that engagement with a personalized virtual platform can induce real learning and enhance farmers’ beliefs and technology choices.